Modeling protein-protein binding is still a massively unsolved problem, mainly because we don't really have the data. Alphafold2 was great but didn't actually 'solve' protein-folding as all input data is from single 'state' X-ray crystallography of the proteins, not 'really' how these proteins behave in the wild. So it's still very, very had to predict what binds to what, which of course is a multi-billion-dollar industry.
I work in a pharma-field and I wish we could easily design molecular binders. We still spend millions every year finding targets that could 'smuggle' our drugs into cells.
Some other players in this field are Boltz Lab and Isomorphic Labs (the Alphafold Google spinoff led by Hasabi). None of them can predict anything complex or 'big', everything is peptide-level. OP's work is another step towards something better.
The most interesting part in the preprint is that they find no matches for their designed binders in the world-write protein database. An open question with protein-designers is whether they just regurgitate training material, which is far easier to test with English-language models.
> very had to predict what binds to what, which of course is a multi-billion-dollar industry.
Do you need to predict when AP-MS is so cheap?
Mapping interaction interfaces is challenging and is where there is attention. I don’t think we’re going to get complexes as a commercial focus outside of receptors with known quaternary structure. The first issue, as you allude to, is the absence of training data, which itself highlights the relative commercial unimportance of such an endeavor.
It's interesting that there are almost no comments on this. This feels like some of the most exciting and impactful fields of the next years. I worked with a cracked researcher that was generating molecules a couple of years ago. She spent most of her time fighting cuda bugs and trying installing packages. I wonder if the ecosystem matured right now. There are people studying cells to see what enters and what exits and engineer how to stop, for example, resources feeding a bad cell. Possibilities feel endless. I am a little worried about side effects, since bio is way more chaotic than silicon, but hopefully AI will help with that level of chaos too.
This was posted on a Saturday night (in the US). A story posted at lunch time on a Tuesday is going to get 100x or maybe 10,000x more views than a story posted on a Saturday night.
It's not that HN readers lack intellectual curiosity or have some character flaw or narrow worldview, it's just that few people are reading and commenting between the late hours of Saturday and early morning of Sunday. It's 6 am Sunday in California as I post this.
I'm sure people will take this the wrong way, but a lot of the people who are on HN and who orbit technology circles in SF are really just not actually intellectually curious people.
They might like to think they are, they might try to pretend they are, but when pushed they're simply not.
Look at all of the groupthink that is perpetuated nonstop while they also proclaim they're creating, investing in, etc. so many unique ideas. Yet year after year it's the same thing in a different color.
What they actually are is interested in money and prestige. So give it a little time and they'll learn enough about biology to try and get some validation from their peers with comments. If money actually pours into bio that is.
I'd go even further: what happens in biology is antithetical to the way software people think.
The HN/YC crowd generally has software brain: https://www.theverge.com/podcast/917029/software-brain-ai-ba..., "when you see the whole world as a series of databases that can be controlled with the structured language of software code". Biology doesn't work like that most of the time, it's squishy and weird and unpredictable, and the models we have of biology (including genomics!) are faulty at best, misleading at worst. I've supervised PhD-students and it takes some time for people's brains to be comfortable with that squishiness, that random behaviour, that 'putting A into the system only rarely produces B and we don't really know why but we do it anyway' view of the world. Software engineers struggle, even abhor that kind of world, which is why you rarely see them being interested in it; and if they work in it, outcomes are sometimes amazing and Nobel Prize worthy, more often nonsense that silently disappears.
> Biology doesn't work like that most of the time, it's squishy and weird and unpredictable, and the models we have of biology (including genomics!) are faulty at best, misleading at worst.
interesting. i came to tech from a molecular biology background and my impression was the opposite. biology is predictable most of the time, but sometimes random and squishy. the trick is that we’re trying to learn why things work predictably and what causes the variations, and that why/how unknown is what is most uncomfortable for people outside of the disciplines.
i’m not fully disagreeing with you because it sounds like you have experiences that inform your perspective. i find it interesting because my own experiences bring me in from the inverse perspective.
The world of uncertainty and the idea that we might not be able to understand everything or control it as much as we'd like.
It seems to me a lot of the modern "tech-bro culture" is trying to control the future and reduce uncertainty: Stop death, merge with the robotic super intelligence, colonize Mars to escape Earth inevitable decay, etc.
I'm still waiting for the startups claiming to reduce entropy or solve the false vacuum decay
Quite frankly, most people on HN are software devs with a wider interest in the world. HN’ers usually-ish comment when they have something insightful to sat, even if the insight is just a humble one.
But I dare to guess that most HN’ers did high school bio and that’s it, so it’s harder to even give a small thoughtful comment on it, so they refrain.
Case in point, I wouldn’t have commented either. But I feel at home here and notice some behavioral patterns. And compared to other fellow devs, I generally am more tuned to tune in on behavioral patterns because of having studied psychology.
I think this is it. I'm a general software engineer and would like to switch to being in this field but something like this is just way over my head. It sounds like a great innovation but I'm lacking the domain knowledge to fully understand it. I've spent several months going through online courses to basically get to a Biology 101 understanding. Getting to the level of understanding something like this seems like it would be a multi-year effort and I don't really know what's the best way to proceed.
all this is part of the new AI for Science effort we are spinning up at Latent Space - all guidance and support would be greatly appreciated as this is a much harder domain to cover than software
It is a nice work, however the domain specific finetuning will always be of higher accuracy prediction. Another thing worth noting is the sequence length used for the training (usually cut to 1024/2048) which is a game changer if left uncut.
I did have a bit of fun myself finetuning esm2 in domain specific bacteria (cause it gives better score) and comparing it to another model (self created) and self created beat it at 25% more accuracy. Then for the 3d structure was coded a 3d protein visualizer hypergraph with the upload file option and visualize instantly the result. 2 days job :)
A similar work is Foundry (https://github.com/RosettaCommons/foundry). While both of them are good, the main issue is that it is not accurate enough at atom level. There are good chances that predicted or designed active site is slightly different from the real structure solved by X-ray, NMR or cryo-electronic microscopy. A side-chain or two may turn the other way so that it changes how the interaction is interpreted. So the tools are good and convenient now. But the design or prediction is often hit-and-miss.
a scientific engine for prediction, design, and discovery that can map proteins across the tree of life, predict their structures, and design new protein binders that function in laboratory experiments.
So, my issue with this is just like in a lot of the other areas of bio we're not able to explore outside the semantics of what is "known." Even a simpler task of just doing proper assembly is plagued by this. De Novo assembly of an alien/novel organism mixed with samples from other alien organisms would be impossible with what we can do today. Even with things that we're familiar we struggle with metagenomic assembly.
The accompanying preprint is interesting: https://www.biorxiv.org/content/10.64898/2026.06.03.729735v1
Modeling protein-protein binding is still a massively unsolved problem, mainly because we don't really have the data. Alphafold2 was great but didn't actually 'solve' protein-folding as all input data is from single 'state' X-ray crystallography of the proteins, not 'really' how these proteins behave in the wild. So it's still very, very had to predict what binds to what, which of course is a multi-billion-dollar industry.
I work in a pharma-field and I wish we could easily design molecular binders. We still spend millions every year finding targets that could 'smuggle' our drugs into cells.
Some other players in this field are Boltz Lab and Isomorphic Labs (the Alphafold Google spinoff led by Hasabi). None of them can predict anything complex or 'big', everything is peptide-level. OP's work is another step towards something better.
The most interesting part in the preprint is that they find no matches for their designed binders in the world-write protein database. An open question with protein-designers is whether they just regurgitate training material, which is far easier to test with English-language models.
> very had to predict what binds to what, which of course is a multi-billion-dollar industry.
Do you need to predict when AP-MS is so cheap?
Mapping interaction interfaces is challenging and is where there is attention. I don’t think we’re going to get complexes as a commercial focus outside of receptors with known quaternary structure. The first issue, as you allude to, is the absence of training data, which itself highlights the relative commercial unimportance of such an endeavor.
> Do you need to predict when AP-MS is so cheap?
Yes, because the expensive part is making the thing.
> None of them can predict anything complex or 'big', everything is peptide-level.
Is this related to the current peptide boom?
It's interesting that there are almost no comments on this. This feels like some of the most exciting and impactful fields of the next years. I worked with a cracked researcher that was generating molecules a couple of years ago. She spent most of her time fighting cuda bugs and trying installing packages. I wonder if the ecosystem matured right now. There are people studying cells to see what enters and what exits and engineer how to stop, for example, resources feeding a bad cell. Possibilities feel endless. I am a little worried about side effects, since bio is way more chaotic than silicon, but hopefully AI will help with that level of chaos too.
This was posted on a Saturday night (in the US). A story posted at lunch time on a Tuesday is going to get 100x or maybe 10,000x more views than a story posted on a Saturday night.
It's not that HN readers lack intellectual curiosity or have some character flaw or narrow worldview, it's just that few people are reading and commenting between the late hours of Saturday and early morning of Sunday. It's 6 am Sunday in California as I post this.
I'm sure people will take this the wrong way, but a lot of the people who are on HN and who orbit technology circles in SF are really just not actually intellectually curious people.
They might like to think they are, they might try to pretend they are, but when pushed they're simply not.
Look at all of the groupthink that is perpetuated nonstop while they also proclaim they're creating, investing in, etc. so many unique ideas. Yet year after year it's the same thing in a different color.
What they actually are is interested in money and prestige. So give it a little time and they'll learn enough about biology to try and get some validation from their peers with comments. If money actually pours into bio that is.
I'd go even further: what happens in biology is antithetical to the way software people think.
The HN/YC crowd generally has software brain: https://www.theverge.com/podcast/917029/software-brain-ai-ba..., "when you see the whole world as a series of databases that can be controlled with the structured language of software code". Biology doesn't work like that most of the time, it's squishy and weird and unpredictable, and the models we have of biology (including genomics!) are faulty at best, misleading at worst. I've supervised PhD-students and it takes some time for people's brains to be comfortable with that squishiness, that random behaviour, that 'putting A into the system only rarely produces B and we don't really know why but we do it anyway' view of the world. Software engineers struggle, even abhor that kind of world, which is why you rarely see them being interested in it; and if they work in it, outcomes are sometimes amazing and Nobel Prize worthy, more often nonsense that silently disappears.
> Biology doesn't work like that most of the time, it's squishy and weird and unpredictable, and the models we have of biology (including genomics!) are faulty at best, misleading at worst.
interesting. i came to tech from a molecular biology background and my impression was the opposite. biology is predictable most of the time, but sometimes random and squishy. the trick is that we’re trying to learn why things work predictably and what causes the variations, and that why/how unknown is what is most uncomfortable for people outside of the disciplines.
i’m not fully disagreeing with you because it sounds like you have experiences that inform your perspective. i find it interesting because my own experiences bring me in from the inverse perspective.
No biologist stays an essentialist for long, that is for sure.
The world of uncertainty and the idea that we might not be able to understand everything or control it as much as we'd like.
It seems to me a lot of the modern "tech-bro culture" is trying to control the future and reduce uncertainty: Stop death, merge with the robotic super intelligence, colonize Mars to escape Earth inevitable decay, etc.
I'm still waiting for the startups claiming to reduce entropy or solve the false vacuum decay
Quite frankly, most people on HN are software devs with a wider interest in the world. HN’ers usually-ish comment when they have something insightful to sat, even if the insight is just a humble one.
But I dare to guess that most HN’ers did high school bio and that’s it, so it’s harder to even give a small thoughtful comment on it, so they refrain.
Case in point, I wouldn’t have commented either. But I feel at home here and notice some behavioral patterns. And compared to other fellow devs, I generally am more tuned to tune in on behavioral patterns because of having studied psychology.
But that’s just my take.
I think this is it. I'm a general software engineer and would like to switch to being in this field but something like this is just way over my head. It sounds like a great innovation but I'm lacking the domain knowledge to fully understand it. I've spent several months going through online courses to basically get to a Biology 101 understanding. Getting to the level of understanding something like this seems like it would be a multi-year effort and I don't really know what's the best way to proceed.
we interviewed Alex Rives, cofounder of EvoScale and Head of Science at BioHub - here https://www.latent.space/p/esmfold2
also 3 paper coauthors walked thru it with us: https://youtu.be/4g1bURdKN0Q
all this is part of the new AI for Science effort we are spinning up at Latent Space - all guidance and support would be greatly appreciated as this is a much harder domain to cover than software
Our mission is to cure or prevent all disease
Okay, now you have my attention.
What's the deal on the company behind it? “Biohub is a 501(c)(3) biomedical research organization...” Nonprofit. Nifty!
This all sounds great, but as we have recently seen with, say OpenAI, there is nonprofit and then there is nonprofit. Anyone know which Biohub is?
Biohub is funded by Zuck and his wife - the full name is "Chan Zuckerberg Biohub"
https://en.wikipedia.org/wiki/Chan_Zuckerberg_Biohub
It is a nice work, however the domain specific finetuning will always be of higher accuracy prediction. Another thing worth noting is the sequence length used for the training (usually cut to 1024/2048) which is a game changer if left uncut.
I did have a bit of fun myself finetuning esm2 in domain specific bacteria (cause it gives better score) and comparing it to another model (self created) and self created beat it at 25% more accuracy. Then for the 3d structure was coded a 3d protein visualizer hypergraph with the upload file option and visualize instantly the result. 2 days job :)
A similar work is Foundry (https://github.com/RosettaCommons/foundry). While both of them are good, the main issue is that it is not accurate enough at atom level. There are good chances that predicted or designed active site is slightly different from the real structure solved by X-ray, NMR or cryo-electronic microscopy. A side-chain or two may turn the other way so that it changes how the interaction is interpreted. So the tools are good and convenient now. But the design or prediction is often hit-and-miss.
But that is also a problem with structures derived from the methods you list. None of them are 100% equivalent to in vivi structures.
Incredible, but also scary if you think about what it may be lowering the barrier of entry to…
> This model is released under the MIT License.
Huh, appears to be actually open source, that's a pleasant surprise. Usually these academic models have some weird license attached to them.